4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
31 * Copyright (c) 2012, 2013, Intel Corporation.
32 * Use is subject to license terms.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/osd-zfs/osd_index.c
40 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
41 * Author: Mike Pershin <tappro@whamcloud.com>
44 #define DEBUG_SUBSYSTEM S_OSD
46 #include <lustre_ver.h>
47 #include <libcfs/libcfs.h>
48 #include <obd_support.h>
49 #include <lustre_net.h>
51 #include <obd_class.h>
52 #include <lustre_disk.h>
53 #include <lustre_fid.h>
55 #include "osd_internal.h"
57 #include <sys/dnode.h>
62 #include <sys/spa_impl.h>
63 #include <sys/zfs_znode.h>
64 #include <sys/dmu_tx.h>
65 #include <sys/dmu_objset.h>
66 #include <sys/dsl_prop.h>
67 #include <sys/sa_impl.h>
70 static inline int osd_object_is_zap(dmu_buf_t *db)
72 dmu_buf_impl_t *dbi = (dmu_buf_impl_t *) db;
78 rc = (dn->dn_type == DMU_OT_DIRECTORY_CONTENTS ||
79 dn->dn_type == DMU_OT_USERGROUP_USED);
85 /* We don't actually have direct access to the zap_hashbits() function
86 * so just pretend like we do for now. If this ever breaks we can look at
88 #define zap_hashbits(zc) 48
91 * | cd (16 bits) | hash (48 bits) |
92 * we need it in other form:
93 * |0| hash (48 bit) | cd (15 bit) |
94 * to be a full 64-bit ordered hash so that Lustre readdir can use it to merge
95 * the readdir hashes from multiple directory stripes uniformly on the client.
96 * Another point is sign bit, the hash range should be in [0, 2^63-1] because
97 * loff_t (for llseek) needs to be a positive value. This means the "cd" field
98 * should only be the low 15 bits.
100 uint64_t osd_zap_cursor_serialize(zap_cursor_t *zc)
102 uint64_t zfs_hash = zap_cursor_serialize(zc) & (~0ULL >> 1);
104 return (zfs_hash >> zap_hashbits(zc)) |
105 (zfs_hash << (63 - zap_hashbits(zc)));
108 void osd_zap_cursor_init_serialized(zap_cursor_t *zc, struct objset *os,
109 uint64_t id, uint64_t dirhash)
111 uint64_t zfs_hash = ((dirhash << zap_hashbits(zc)) & (~0ULL >> 1)) |
112 (dirhash >> (63 - zap_hashbits(zc)));
114 zap_cursor_init_serialized(zc, os, id, zfs_hash);
117 int osd_zap_cursor_init(zap_cursor_t **zc, struct objset *os,
118 uint64_t id, uint64_t dirhash)
123 if (unlikely(t == NULL))
126 osd_zap_cursor_init_serialized(t, os, id, dirhash);
132 void osd_zap_cursor_fini(zap_cursor_t *zc)
138 static inline void osd_obj_cursor_init_serialized(zap_cursor_t *zc,
139 struct osd_object *o,
142 struct osd_device *d = osd_obj2dev(o);
143 zap_cursor_init_serialized(zc, d->od_os, o->oo_db->db_object, dirhash);
146 static inline int osd_obj_cursor_init(zap_cursor_t **zc, struct osd_object *o,
149 struct osd_device *d = osd_obj2dev(o);
150 return osd_zap_cursor_init(zc, d->od_os, o->oo_db->db_object, dirhash);
153 static struct dt_it *osd_index_it_init(const struct lu_env *env,
154 struct dt_object *dt,
156 struct lustre_capa *capa)
158 struct osd_thread_info *info = osd_oti_get(env);
159 struct osd_zap_it *it;
160 struct osd_object *obj = osd_dt_obj(dt);
161 struct lu_object *lo = &dt->do_lu;
165 /* XXX: check capa ? */
167 LASSERT(lu_object_exists(lo));
169 LASSERT(osd_object_is_zap(obj->oo_db));
172 it = &info->oti_it_zap;
174 rc = osd_obj_cursor_init(&it->ozi_zc, obj, 0);
183 RETURN((struct dt_it *)it);
186 static void osd_index_it_fini(const struct lu_env *env, struct dt_it *di)
188 struct osd_zap_it *it = (struct osd_zap_it *)di;
189 struct osd_object *obj;
193 LASSERT(it->ozi_obj);
197 osd_zap_cursor_fini(it->ozi_zc);
198 lu_object_put(env, &obj->oo_dt.do_lu);
204 static void osd_index_it_put(const struct lu_env *env, struct dt_it *di)
206 /* PBS: do nothing : ref are incremented at retrive and decreamented
210 static inline void osd_it_append_attrs(struct lu_dirent *ent, __u32 attr,
213 const unsigned align = sizeof(struct luda_type) - 1;
214 struct luda_type *lt;
216 /* check if file type is required */
217 if (attr & LUDA_TYPE) {
218 len = (len + align) & ~align;
220 lt = (void *)ent->lde_name + len;
221 lt->lt_type = cpu_to_le16(DTTOIF(type));
222 ent->lde_attrs |= LUDA_TYPE;
225 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
229 * as we don't know FID, we can't use LU object, so this function
230 * partially duplicate __osd_xattr_get() which is built around
231 * LU-object and uses it to cache data like regular EA dnode, etc
233 static int osd_find_parent_by_dnode(const struct lu_env *env,
237 struct osd_device *osd = osd_obj2dev(osd_dt_obj(o));
238 struct lustre_mdt_attrs *lma;
241 nvlist_t *nvbuf = NULL;
247 /* first of all, get parent dnode from own attributes */
248 LASSERT(osd_dt_obj(o)->oo_db);
249 rc = -sa_handle_get(osd->od_os, osd_dt_obj(o)->oo_db->db_object,
250 NULL, SA_HDL_PRIVATE, &sa_hdl);
254 dnode = ZFS_NO_OBJECT;
255 rc = -sa_lookup(sa_hdl, SA_ZPL_PARENT(osd), &dnode, 8);
256 sa_handle_destroy(sa_hdl);
260 /* now get EA buffer */
261 rc = __osd_xattr_load(osd, dnode, &nvbuf);
265 /* XXX: if we get that far.. should we cache the result? */
267 /* try to find LMA attribute */
268 LASSERT(nvbuf != NULL);
269 rc = -nvlist_lookup_byte_array(nvbuf, XATTR_NAME_LMA, &value, &size);
270 if (rc == 0 && size >= sizeof(*lma)) {
271 lma = (struct lustre_mdt_attrs *)value;
272 lustre_lma_swab(lma);
273 *fid = lma->lma_self_fid;
278 /* no LMA attribute in SA, let's try regular EA */
280 /* first of all, get parent dnode storing regular EA */
281 rc = -sa_handle_get(osd->od_os, dnode, NULL, SA_HDL_PRIVATE, &sa_hdl);
285 dnode = ZFS_NO_OBJECT;
286 rc = -sa_lookup(sa_hdl, SA_ZPL_XATTR(osd), &dnode, 8);
287 sa_handle_destroy(sa_hdl);
291 CLASSERT(sizeof(*lma) <= sizeof(osd_oti_get(env)->oti_buf));
292 buf.lb_buf = osd_oti_get(env)->oti_buf;
293 buf.lb_len = sizeof(osd_oti_get(env)->oti_buf);
295 /* now try to find LMA */
296 rc = __osd_xattr_get_large(env, osd, dnode, &buf,
297 XATTR_NAME_LMA, &size);
298 if (rc == 0 && size >= sizeof(*lma)) {
300 lustre_lma_swab(lma);
301 *fid = lma->lma_self_fid;
306 GOTO(out, rc = -EIO);
315 static int osd_find_parent_fid(const struct lu_env *env, struct dt_object *o,
318 struct link_ea_header *leh;
319 struct link_ea_entry *lee;
324 buf.lb_buf = osd_oti_get(env)->oti_buf;
325 buf.lb_len = sizeof(osd_oti_get(env)->oti_buf);
327 rc = osd_xattr_get(env, o, &buf, XATTR_NAME_LINK, BYPASS_CAPA);
329 rc = osd_xattr_get(env, o, &LU_BUF_NULL,
330 XATTR_NAME_LINK, BYPASS_CAPA);
334 OBD_ALLOC(buf.lb_buf, rc);
335 if (buf.lb_buf == NULL)
338 rc = osd_xattr_get(env, o, &buf, XATTR_NAME_LINK, BYPASS_CAPA);
342 if (rc < sizeof(*leh) + sizeof(*lee))
343 GOTO(out, rc = -EINVAL);
346 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
347 leh->leh_magic = LINK_EA_MAGIC;
348 leh->leh_reccount = __swab32(leh->leh_reccount);
349 leh->leh_len = __swab64(leh->leh_len);
351 if (leh->leh_magic != LINK_EA_MAGIC)
352 GOTO(out, rc = -EINVAL);
353 if (leh->leh_reccount == 0)
354 GOTO(out, rc = -ENODATA);
356 lee = (struct link_ea_entry *)(leh + 1);
357 fid_be_to_cpu(fid, (const struct lu_fid *)&lee->lee_parent_fid);
361 if (buf.lb_buf != osd_oti_get(env)->oti_buf)
362 OBD_FREE(buf.lb_buf, buf.lb_len);
365 /* this block can be enabled for additional verification
366 * it's trying to match FID from LinkEA vs. FID from LMA */
370 rc2 = osd_find_parent_by_dnode(env, o, &fid2);
372 if (lu_fid_eq(fid, &fid2) == 0)
373 CERROR("wrong parent: "DFID" != "DFID"\n",
374 PFID(fid), PFID(&fid2));
378 /* no LinkEA is found, let's try to find the fid in parent's LMA */
379 if (unlikely(rc != 0))
380 rc = osd_find_parent_by_dnode(env, o, fid);
385 static int osd_dir_lookup(const struct lu_env *env, struct dt_object *dt,
386 struct dt_rec *rec, const struct dt_key *key,
387 struct lustre_capa *capa)
389 struct osd_thread_info *oti = osd_oti_get(env);
390 struct osd_object *obj = osd_dt_obj(dt);
391 struct osd_device *osd = osd_obj2dev(obj);
392 char *name = (char *)key;
396 LASSERT(osd_object_is_zap(obj->oo_db));
398 if (name[0] == '.') {
400 const struct lu_fid *f = lu_object_fid(&dt->do_lu);
401 memcpy(rec, f, sizeof(*f));
403 } else if (name[1] == '.' && name[2] == 0) {
404 rc = osd_find_parent_fid(env, dt, (struct lu_fid *)rec);
405 RETURN(rc == 0 ? 1 : rc);
409 rc = -zap_lookup(osd->od_os, obj->oo_db->db_object,
410 (char *)key, 8, sizeof(oti->oti_zde) / 8,
411 (void *)&oti->oti_zde);
412 memcpy(rec, &oti->oti_zde.lzd_fid, sizeof(struct lu_fid));
414 RETURN(rc == 0 ? 1 : rc);
417 static int osd_declare_dir_insert(const struct lu_env *env,
418 struct dt_object *dt,
419 const struct dt_rec *rec,
420 const struct dt_key *key,
423 struct osd_object *obj = osd_dt_obj(dt);
424 struct osd_thandle *oh;
428 oh = container_of0(th, struct osd_thandle, ot_super);
431 LASSERT(osd_object_is_zap(obj->oo_db));
433 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_db->db_object);
434 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, (char *)key);
440 * Find the osd object for given fid.
442 * \param fid need to find the osd object having this fid
444 * \retval osd_object on success
445 * \retval -ve on error
447 struct osd_object *osd_object_find(const struct lu_env *env,
448 struct dt_object *dt,
449 const struct lu_fid *fid)
451 struct lu_device *ludev = dt->do_lu.lo_dev;
452 struct osd_object *child = NULL;
453 struct lu_object *luch;
454 struct lu_object *lo;
457 * at this point topdev might not exist yet
458 * (i.e. MGS is preparing profiles). so we can
459 * not rely on topdev and instead lookup with
460 * our device passed as topdev. this can't work
461 * if the object isn't cached yet (as osd doesn't
462 * allocate lu_header). IOW, the object must be
463 * in the cache, otherwise lu_object_alloc() crashes
466 luch = lu_object_find_at(env, ludev, fid, NULL);
470 if (lu_object_exists(luch)) {
471 lo = lu_object_locate(luch->lo_header, ludev->ld_type);
475 LU_OBJECT_DEBUG(D_ERROR, env, luch,
476 "%s: object can't be located "DFID"\n",
477 osd_dev(ludev)->od_svname, PFID(fid));
480 lu_object_put(env, luch);
481 CERROR("%s: Unable to get osd_object "DFID"\n",
482 osd_dev(ludev)->od_svname, PFID(fid));
483 child = ERR_PTR(-ENOENT);
486 LU_OBJECT_DEBUG(D_ERROR, env, luch,
487 "%s: lu_object does not exists "DFID"\n",
488 osd_dev(ludev)->od_svname, PFID(fid));
489 lu_object_put(env, luch);
490 child = ERR_PTR(-ENOENT);
497 * Put the osd object once done with it.
499 * \param obj osd object that needs to be put
501 static inline void osd_object_put(const struct lu_env *env,
502 struct osd_object *obj)
504 lu_object_put(env, &obj->oo_dt.do_lu);
507 static int osd_seq_exists(const struct lu_env *env, struct osd_device *osd,
510 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
511 struct seq_server_site *ss = osd_seq_site(osd);
516 LASSERT(ss->ss_server_fld != NULL);
518 rc = osd_fld_lookup(env, osd, seq, range);
520 CERROR("%s: Can not lookup fld for "LPX64"\n",
525 RETURN(ss->ss_node_id == range->lsr_index);
528 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
529 const struct lu_fid *fid)
531 struct seq_server_site *ss = osd_seq_site(osd);
534 /* FID seqs not in FLDB, must be local seq */
535 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
538 /* If FLD is not being initialized yet, it only happens during the
539 * initialization, likely during mgs initialization, and we assume
540 * this is local FID. */
541 if (ss == NULL || ss->ss_server_fld == NULL)
544 /* Only check the local FLDB here */
545 if (osd_seq_exists(env, osd, fid_seq(fid)))
552 * Inserts (key, value) pair in \a directory object.
554 * \param dt osd index object
555 * \param key key for index
556 * \param rec record reference
557 * \param th transaction handler
558 * \param capa capability descriptor
559 * \param ignore_quota update should not affect quota
562 * \retval -ve failure
564 static int osd_dir_insert(const struct lu_env *env, struct dt_object *dt,
565 const struct dt_rec *rec, const struct dt_key *key,
566 struct thandle *th, struct lustre_capa *capa,
569 struct osd_thread_info *oti = osd_oti_get(env);
570 struct osd_object *parent = osd_dt_obj(dt);
571 struct osd_device *osd = osd_obj2dev(parent);
572 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
573 const struct lu_fid *fid = rec1->rec_fid;
574 struct osd_thandle *oh;
575 struct osd_object *child = NULL;
577 char *name = (char *)key;
581 LASSERT(parent->oo_db);
582 LASSERT(osd_object_is_zap(parent->oo_db));
584 LASSERT(dt_object_exists(dt));
585 LASSERT(osd_invariant(parent));
588 oh = container_of0(th, struct osd_thandle, ot_super);
590 rc = osd_remote_fid(env, osd, fid);
592 CERROR("%s: Can not find object "DFID": rc = %d\n",
593 osd->od_svname, PFID(fid), rc);
597 if (unlikely(rc == 1)) {
598 /* Insert remote entry */
599 memset(&oti->oti_zde.lzd_reg, 0, sizeof(oti->oti_zde.lzd_reg));
600 oti->oti_zde.lzd_reg.zde_type = IFTODT(rec1->rec_type & S_IFMT);
603 * To simulate old Orion setups with ./.. stored in the
606 /* Insert local entry */
607 child = osd_object_find(env, dt, fid);
609 RETURN(PTR_ERR(child));
611 LASSERT(child->oo_db);
612 if (name[0] == '.') {
614 /* do not store ".", instead generate it
615 * during iteration */
617 } else if (name[1] == '.' && name[2] == 0) {
618 /* update parent dnode in the child.
619 * later it will be used to generate ".." */
620 rc = osd_object_sa_update(parent,
622 &child->oo_db->db_object,
627 CLASSERT(sizeof(oti->oti_zde.lzd_reg) == 8);
628 CLASSERT(sizeof(oti->oti_zde) % 8 == 0);
629 attr = child->oo_dt.do_lu.lo_header ->loh_attr;
630 oti->oti_zde.lzd_reg.zde_type = IFTODT(attr & S_IFMT);
631 oti->oti_zde.lzd_reg.zde_dnode = child->oo_db->db_object;
634 oti->oti_zde.lzd_fid = *fid;
635 /* Insert (key,oid) into ZAP */
636 rc = -zap_add(osd->od_os, parent->oo_db->db_object,
637 (char *)key, 8, sizeof(oti->oti_zde) / 8,
638 (void *)&oti->oti_zde, oh->ot_tx);
642 osd_object_put(env, child);
647 static int osd_declare_dir_delete(const struct lu_env *env,
648 struct dt_object *dt,
649 const struct dt_key *key,
652 struct osd_object *obj = osd_dt_obj(dt);
653 struct osd_thandle *oh;
656 LASSERT(dt_object_exists(dt));
657 LASSERT(osd_invariant(obj));
660 oh = container_of0(th, struct osd_thandle, ot_super);
663 LASSERT(osd_object_is_zap(obj->oo_db));
665 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, (char *)key);
670 static int osd_dir_delete(const struct lu_env *env, struct dt_object *dt,
671 const struct dt_key *key, struct thandle *th,
672 struct lustre_capa *capa)
674 struct osd_object *obj = osd_dt_obj(dt);
675 struct osd_device *osd = osd_obj2dev(obj);
676 struct osd_thandle *oh;
677 dmu_buf_t *zap_db = obj->oo_db;
678 char *name = (char *)key;
683 LASSERT(osd_object_is_zap(obj->oo_db));
686 oh = container_of0(th, struct osd_thandle, ot_super);
689 * In Orion . and .. were stored in the directory (not generated upon
690 * request as now). we preserve them for backward compatibility
692 if (name[0] == '.') {
695 } else if (name[1] == '.' && name[2] == 0) {
700 /* Remove key from the ZAP */
701 rc = -zap_remove(osd->od_os, zap_db->db_object,
702 (char *) key, oh->ot_tx);
704 if (unlikely(rc && rc != -ENOENT))
705 CERROR("%s: zap_remove failed: rc = %d\n", osd->od_svname, rc);
710 static struct dt_it *osd_dir_it_init(const struct lu_env *env,
711 struct dt_object *dt,
713 struct lustre_capa *capa)
715 struct osd_zap_it *it;
717 it = (struct osd_zap_it *)osd_index_it_init(env, dt, unused, capa);
721 RETURN((struct dt_it *)it);
725 * Move Iterator to record specified by \a key
727 * \param di osd iterator
728 * \param key key for index
730 * \retval +ve di points to record with least key not larger than key
731 * \retval 0 di points to exact matched key
732 * \retval -ve failure
734 static int osd_dir_it_get(const struct lu_env *env,
735 struct dt_it *di, const struct dt_key *key)
737 struct osd_zap_it *it = (struct osd_zap_it *)di;
738 struct osd_object *obj = it->ozi_obj;
739 char *name = (char *)key;
746 /* reset the cursor */
747 zap_cursor_fini(it->ozi_zc);
748 osd_obj_cursor_init_serialized(it->ozi_zc, obj, 0);
750 /* XXX: implementation of the API is broken at the moment */
751 LASSERT(((const char *)key)[0] == 0);
758 if (name[0] == '.') {
762 } else if (name[1] == '.' && name[2] == 0) {
768 /* neither . nor .. - some real record */
776 static void osd_dir_it_put(const struct lu_env *env, struct dt_it *di)
778 /* PBS: do nothing : ref are incremented at retrive and decreamented
783 * in Orion . and .. were stored in the directory, while ZPL
784 * and current osd-zfs generate them up on request. so, we
785 * need to ignore previously stored . and ..
787 static int osd_index_retrieve_skip_dots(struct osd_zap_it *it,
793 rc = -zap_cursor_retrieve(it->ozi_zc, za);
796 if (unlikely(rc == 0 && za->za_name[0] == '.')) {
797 if (za->za_name[1] == 0) {
799 } else if (za->za_name[1] == '.' &&
800 za->za_name[2] == 0) {
804 zap_cursor_advance(it->ozi_zc);
806 } while (unlikely(rc == 0 && isdot));
812 * to load a directory entry at a time and stored it in
813 * iterator's in-memory data structure.
815 * \param di, struct osd_it_ea, iterator's in memory structure
817 * \retval +ve, iterator reached to end
818 * \retval 0, iterator not reached to end
819 * \retval -ve, on error
821 static int osd_dir_it_next(const struct lu_env *env, struct dt_it *di)
823 struct osd_zap_it *it = (struct osd_zap_it *)di;
824 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
827 /* temp. storage should be enough for any key supported by ZFS */
828 CLASSERT(sizeof(za->za_name) <= sizeof(it->ozi_name));
831 * the first ->next() moves the cursor to .
832 * the second ->next() moves the cursor to ..
833 * then we get to the real records and have to verify any exist
835 if (it->ozi_pos <= 2) {
841 zap_cursor_advance(it->ozi_zc);
844 * According to current API we need to return error if its last entry.
845 * zap_cursor_advance() does not return any value. So we need to call
846 * retrieve to check if there is any record. We should make
847 * changes to Iterator API to not return status for this API
849 rc = osd_index_retrieve_skip_dots(it, za);
851 if (rc == -ENOENT) /* end of dir */
857 static struct dt_key *osd_dir_it_key(const struct lu_env *env,
858 const struct dt_it *di)
860 struct osd_zap_it *it = (struct osd_zap_it *)di;
861 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
865 if (it->ozi_pos <= 1) {
867 RETURN((struct dt_key *)".");
868 } else if (it->ozi_pos == 2) {
869 RETURN((struct dt_key *)"..");
872 if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)))
875 strcpy(it->ozi_name, za->za_name);
877 RETURN((struct dt_key *)it->ozi_name);
880 static int osd_dir_it_key_size(const struct lu_env *env, const struct dt_it *di)
882 struct osd_zap_it *it = (struct osd_zap_it *)di;
883 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
887 if (it->ozi_pos <= 1) {
890 } else if (it->ozi_pos == 2) {
894 if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)) == 0)
895 rc = strlen(za->za_name);
900 static int osd_dir_it_rec(const struct lu_env *env, const struct dt_it *di,
901 struct dt_rec *dtrec, __u32 attr)
903 struct osd_zap_it *it = (struct osd_zap_it *)di;
904 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
905 struct luz_direntry *zde = &osd_oti_get(env)->oti_zde;
906 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
910 if (it->ozi_pos <= 1) {
911 lde->lde_hash = cpu_to_le64(1);
912 strcpy(lde->lde_name, ".");
913 lde->lde_namelen = cpu_to_le16(1);
914 lde->lde_fid = *lu_object_fid(&it->ozi_obj->oo_dt.do_lu);
915 lde->lde_attrs = LUDA_FID;
916 /* append lustre attributes */
917 osd_it_append_attrs(lde, attr, 1, IFTODT(S_IFDIR));
918 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(1, attr));
922 } else if (it->ozi_pos == 2) {
923 lde->lde_hash = cpu_to_le64(2);
924 strcpy(lde->lde_name, "..");
925 lde->lde_namelen = cpu_to_le16(2);
926 lde->lde_attrs = LUDA_FID;
927 /* append lustre attributes */
928 osd_it_append_attrs(lde, attr, 2, IFTODT(S_IFDIR));
929 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(2, attr));
930 rc = osd_find_parent_fid(env, &it->ozi_obj->oo_dt, &lde->lde_fid);
932 * early Orion code was not setting LinkEA, so it's possible
933 * some setups still have objects with no LinkEA set.
934 * but at that time .. was a real record in the directory
935 * so we should try to lookup .. in ZAP
943 rc = -zap_cursor_retrieve(it->ozi_zc, za);
944 if (unlikely(rc != 0))
947 lde->lde_hash = cpu_to_le64(osd_zap_cursor_serialize(it->ozi_zc));
948 namelen = strlen(za->za_name);
949 if (namelen > NAME_MAX)
950 GOTO(out, rc = -EOVERFLOW);
951 strcpy(lde->lde_name, za->za_name);
952 lde->lde_namelen = cpu_to_le16(namelen);
954 if (za->za_integer_length != 8 || za->za_num_integers < 3) {
955 CERROR("%s: unsupported direntry format: %d %d\n",
956 osd_obj2dev(it->ozi_obj)->od_svname,
957 za->za_integer_length, (int)za->za_num_integers);
959 GOTO(out, rc = -EIO);
962 rc = -zap_lookup(it->ozi_zc->zc_objset, it->ozi_zc->zc_zapobj,
963 za->za_name, za->za_integer_length, 3, zde);
967 lde->lde_fid = zde->lzd_fid;
968 lde->lde_attrs = LUDA_FID;
970 /* append lustre attributes */
971 osd_it_append_attrs(lde, attr, namelen, zde->lzd_reg.zde_type);
973 lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
979 static int osd_dir_it_rec_size(const struct lu_env *env, const struct dt_it *di,
982 struct osd_zap_it *it = (struct osd_zap_it *)di;
983 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
987 if (it->ozi_pos <= 1)
989 else if (it->ozi_pos == 2)
993 rc = lu_dirent_calc_size(namelen, attr);
997 rc = -zap_cursor_retrieve(it->ozi_zc, za);
998 if (unlikely(rc != 0))
1001 if (za->za_integer_length != 8 || za->za_num_integers < 3) {
1002 CERROR("%s: unsupported direntry format: %d %d\n",
1003 osd_obj2dev(it->ozi_obj)->od_svname,
1004 za->za_integer_length, (int)za->za_num_integers);
1008 namelen = strlen(za->za_name);
1009 if (namelen > NAME_MAX)
1012 rc = lu_dirent_calc_size(namelen, attr);
1017 static __u64 osd_dir_it_store(const struct lu_env *env, const struct dt_it *di)
1019 struct osd_zap_it *it = (struct osd_zap_it *)di;
1023 if (it->ozi_pos <= 2)
1026 pos = osd_zap_cursor_serialize(it->ozi_zc);
1033 * rc == 0 -> end of directory.
1034 * rc > 0 -> ok, proceed.
1035 * rc < 0 -> error. ( EOVERFLOW can be masked.)
1037 static int osd_dir_it_load(const struct lu_env *env,
1038 const struct dt_it *di, __u64 hash)
1040 struct osd_zap_it *it = (struct osd_zap_it *)di;
1041 struct osd_object *obj = it->ozi_obj;
1042 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1046 /* reset the cursor */
1047 zap_cursor_fini(it->ozi_zc);
1048 osd_obj_cursor_init_serialized(it->ozi_zc, obj, hash);
1055 /* to return whether the end has been reached */
1056 rc = osd_index_retrieve_skip_dots(it, za);
1059 else if (rc == -ENOENT)
1066 static struct dt_index_operations osd_dir_ops = {
1067 .dio_lookup = osd_dir_lookup,
1068 .dio_declare_insert = osd_declare_dir_insert,
1069 .dio_insert = osd_dir_insert,
1070 .dio_declare_delete = osd_declare_dir_delete,
1071 .dio_delete = osd_dir_delete,
1073 .init = osd_dir_it_init,
1074 .fini = osd_index_it_fini,
1075 .get = osd_dir_it_get,
1076 .put = osd_dir_it_put,
1077 .next = osd_dir_it_next,
1078 .key = osd_dir_it_key,
1079 .key_size = osd_dir_it_key_size,
1080 .rec = osd_dir_it_rec,
1081 .rec_size = osd_dir_it_rec_size,
1082 .store = osd_dir_it_store,
1083 .load = osd_dir_it_load
1088 * Primitives for index files using binary keys.
1091 /* key integer_size is 8 */
1092 static int osd_prepare_key_uint64(struct osd_object *o, __u64 *dst,
1093 const struct dt_key *src)
1100 /* align keysize to 64bit */
1101 size = (o->oo_keysize + sizeof(__u64) - 1) / sizeof(__u64);
1102 size *= sizeof(__u64);
1104 LASSERT(size <= MAXNAMELEN);
1106 if (unlikely(size > o->oo_keysize))
1107 memset(dst + o->oo_keysize, 0, size - o->oo_keysize);
1108 memcpy(dst, (const char *)src, o->oo_keysize);
1110 return (size/sizeof(__u64));
1113 static int osd_index_lookup(const struct lu_env *env, struct dt_object *dt,
1114 struct dt_rec *rec, const struct dt_key *key,
1115 struct lustre_capa *capa)
1117 struct osd_object *obj = osd_dt_obj(dt);
1118 struct osd_device *osd = osd_obj2dev(obj);
1119 __u64 *k = osd_oti_get(env)->oti_key64;
1123 rc = osd_prepare_key_uint64(obj, k, key);
1125 rc = -zap_lookup_uint64(osd->od_os, obj->oo_db->db_object,
1126 k, rc, obj->oo_recusize, obj->oo_recsize,
1128 RETURN(rc == 0 ? 1 : rc);
1131 static int osd_declare_index_insert(const struct lu_env *env,
1132 struct dt_object *dt,
1133 const struct dt_rec *rec,
1134 const struct dt_key *key,
1137 struct osd_object *obj = osd_dt_obj(dt);
1138 struct osd_thandle *oh;
1141 LASSERT(th != NULL);
1142 oh = container_of0(th, struct osd_thandle, ot_super);
1144 LASSERT(obj->oo_db);
1146 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_db->db_object);
1148 /* It is not clear what API should be used for binary keys, so we pass
1149 * a null name which has the side effect of over-reserving space,
1150 * accounting for the worst case. See zap_count_write() */
1151 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, NULL);
1156 static int osd_index_insert(const struct lu_env *env, struct dt_object *dt,
1157 const struct dt_rec *rec, const struct dt_key *key,
1158 struct thandle *th, struct lustre_capa *capa,
1161 struct osd_object *obj = osd_dt_obj(dt);
1162 struct osd_device *osd = osd_obj2dev(obj);
1163 struct osd_thandle *oh;
1164 __u64 *k = osd_oti_get(env)->oti_key64;
1168 LASSERT(obj->oo_db);
1169 LASSERT(dt_object_exists(dt));
1170 LASSERT(osd_invariant(obj));
1171 LASSERT(th != NULL);
1173 oh = container_of0(th, struct osd_thandle, ot_super);
1175 rc = osd_prepare_key_uint64(obj, k, key);
1177 /* Insert (key,oid) into ZAP */
1178 rc = -zap_add_uint64(osd->od_os, obj->oo_db->db_object,
1179 k, rc, obj->oo_recusize, obj->oo_recsize,
1180 (void *)rec, oh->ot_tx);
1184 static int osd_declare_index_delete(const struct lu_env *env,
1185 struct dt_object *dt,
1186 const struct dt_key *key,
1189 struct osd_object *obj = osd_dt_obj(dt);
1190 struct osd_thandle *oh;
1193 LASSERT(dt_object_exists(dt));
1194 LASSERT(osd_invariant(obj));
1195 LASSERT(th != NULL);
1196 LASSERT(obj->oo_db);
1198 oh = container_of0(th, struct osd_thandle, ot_super);
1199 dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, NULL);
1204 static int osd_index_delete(const struct lu_env *env, struct dt_object *dt,
1205 const struct dt_key *key, struct thandle *th,
1206 struct lustre_capa *capa)
1208 struct osd_object *obj = osd_dt_obj(dt);
1209 struct osd_device *osd = osd_obj2dev(obj);
1210 struct osd_thandle *oh;
1211 __u64 *k = osd_oti_get(env)->oti_key64;
1215 LASSERT(obj->oo_db);
1216 LASSERT(th != NULL);
1217 oh = container_of0(th, struct osd_thandle, ot_super);
1219 rc = osd_prepare_key_uint64(obj, k, key);
1221 /* Remove binary key from the ZAP */
1222 rc = -zap_remove_uint64(osd->od_os, obj->oo_db->db_object,
1227 static int osd_index_it_get(const struct lu_env *env, struct dt_it *di,
1228 const struct dt_key *key)
1230 struct osd_zap_it *it = (struct osd_zap_it *)di;
1231 struct osd_object *obj = it->ozi_obj;
1232 struct osd_device *osd = osd_obj2dev(obj);
1236 LASSERT(it->ozi_zc);
1239 * XXX: we need a binary version of zap_cursor_move_to_key()
1240 * to implement this API */
1241 if (*((const __u64 *)key) != 0)
1242 CERROR("NOT IMPLEMETED YET (move to "LPX64")\n",
1245 zap_cursor_fini(it->ozi_zc);
1246 zap_cursor_init(it->ozi_zc, osd->od_os, obj->oo_db->db_object);
1252 static int osd_index_it_next(const struct lu_env *env, struct dt_it *di)
1254 struct osd_zap_it *it = (struct osd_zap_it *)di;
1255 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1259 if (it->ozi_reset == 0)
1260 zap_cursor_advance(it->ozi_zc);
1264 * According to current API we need to return error if it's last entry.
1265 * zap_cursor_advance() does not return any value. So we need to call
1266 * retrieve to check if there is any record. We should make
1267 * changes to Iterator API to not return status for this API
1269 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1276 static struct dt_key *osd_index_it_key(const struct lu_env *env,
1277 const struct dt_it *di)
1279 struct osd_zap_it *it = (struct osd_zap_it *)di;
1280 struct osd_object *obj = it->ozi_obj;
1281 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1286 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1288 RETURN(ERR_PTR(rc));
1290 /* the binary key is stored in the name */
1291 memcpy(&it->ozi_key, za->za_name, obj->oo_keysize);
1293 RETURN((struct dt_key *)&it->ozi_key);
1296 static int osd_index_it_key_size(const struct lu_env *env,
1297 const struct dt_it *di)
1299 struct osd_zap_it *it = (struct osd_zap_it *)di;
1300 struct osd_object *obj = it->ozi_obj;
1301 RETURN(obj->oo_keysize);
1304 static int osd_index_it_rec(const struct lu_env *env, const struct dt_it *di,
1305 struct dt_rec *rec, __u32 attr)
1307 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1308 struct osd_zap_it *it = (struct osd_zap_it *)di;
1309 struct osd_object *obj = it->ozi_obj;
1310 struct osd_device *osd = osd_obj2dev(obj);
1311 __u64 *k = osd_oti_get(env)->oti_key64;
1316 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1320 rc = osd_prepare_key_uint64(obj, k, (const struct dt_key *)za->za_name);
1322 rc = -zap_lookup_uint64(osd->od_os, obj->oo_db->db_object,
1323 k, rc, obj->oo_recusize, obj->oo_recsize,
1328 static __u64 osd_index_it_store(const struct lu_env *env,
1329 const struct dt_it *di)
1331 struct osd_zap_it *it = (struct osd_zap_it *)di;
1334 RETURN((__u64)zap_cursor_serialize(it->ozi_zc));
1337 static int osd_index_it_load(const struct lu_env *env, const struct dt_it *di,
1340 struct osd_zap_it *it = (struct osd_zap_it *)di;
1341 struct osd_object *obj = it->ozi_obj;
1342 struct osd_device *osd = osd_obj2dev(obj);
1343 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1347 /* reset the cursor */
1348 zap_cursor_fini(it->ozi_zc);
1349 zap_cursor_init_serialized(it->ozi_zc, osd->od_os,
1350 obj->oo_db->db_object, hash);
1353 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1356 else if (rc == -ENOENT)
1362 static struct dt_index_operations osd_index_ops = {
1363 .dio_lookup = osd_index_lookup,
1364 .dio_declare_insert = osd_declare_index_insert,
1365 .dio_insert = osd_index_insert,
1366 .dio_declare_delete = osd_declare_index_delete,
1367 .dio_delete = osd_index_delete,
1369 .init = osd_index_it_init,
1370 .fini = osd_index_it_fini,
1371 .get = osd_index_it_get,
1372 .put = osd_index_it_put,
1373 .next = osd_index_it_next,
1374 .key = osd_index_it_key,
1375 .key_size = osd_index_it_key_size,
1376 .rec = osd_index_it_rec,
1377 .store = osd_index_it_store,
1378 .load = osd_index_it_load
1382 struct osd_metadnode_it {
1383 struct osd_device *mit_dev;
1385 struct lu_fid mit_fid;
1387 __u64 mit_prefetched_dnode;
1390 static struct dt_it *osd_zfs_otable_it_init(const struct lu_env *env,
1391 struct dt_object *dt, __u32 attr,
1392 struct lustre_capa *capa)
1394 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1395 struct osd_metadnode_it *it;
1399 if (unlikely(it == NULL))
1400 RETURN(ERR_PTR(-ENOMEM));
1404 /* XXX: dmu_object_next() does NOT find dnodes allocated
1405 * in the current non-committed txg, so we force txg
1406 * commit to find all existing dnodes ... */
1407 txg_wait_synced(dmu_objset_pool(dev->od_os), 0ULL);
1409 RETURN((struct dt_it *)it);
1412 static void osd_zfs_otable_it_fini(const struct lu_env *env, struct dt_it *di)
1414 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1419 static int osd_zfs_otable_it_get(const struct lu_env *env,
1420 struct dt_it *di, const struct dt_key *key)
1425 static void osd_zfs_otable_it_put(const struct lu_env *env, struct dt_it *di)
1429 #define OTABLE_PREFETCH 256
1431 static void osd_zfs_otable_prefetch(const struct lu_env *env,
1432 struct osd_metadnode_it *it)
1434 struct osd_device *dev = it->mit_dev;
1437 /* can go negative on the very first access to the iterator
1438 * or if some non-Lustre objects were found */
1439 if (unlikely(it->mit_prefetched < 0))
1440 it->mit_prefetched = 0;
1442 if (it->mit_prefetched >= (OTABLE_PREFETCH >> 1))
1445 if (it->mit_prefetched_dnode == 0)
1446 it->mit_prefetched_dnode = it->mit_pos;
1448 while (it->mit_prefetched < OTABLE_PREFETCH) {
1449 rc = -dmu_object_next(dev->od_os, &it->mit_prefetched_dnode,
1451 if (unlikely(rc != 0))
1454 /* dmu_prefetch() was exported in 0.6.2, if you use with
1455 * an older release, just comment it out - this is an
1457 dmu_prefetch(dev->od_os, it->mit_prefetched_dnode, 0, 0);
1459 it->mit_prefetched++;
1463 static int osd_zfs_otable_it_next(const struct lu_env *env, struct dt_it *di)
1465 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1466 struct lustre_mdt_attrs *lma;
1467 struct osd_device *dev = it->mit_dev;
1468 nvlist_t *nvbuf = NULL;
1473 memset(&it->mit_fid, 0, sizeof(it->mit_fid));
1475 dnode = it->mit_pos;
1477 rc = -dmu_object_next(dev->od_os, &it->mit_pos, B_FALSE, 0);
1478 if (unlikely(rc != 0))
1480 it->mit_prefetched--;
1482 /* LMA is required for this to be a Lustre object.
1483 * If there is no xattr skip it. */
1484 rc = __osd_xattr_load(dev, it->mit_pos, &nvbuf);
1485 if (unlikely(rc != 0))
1488 LASSERT(nvbuf != NULL);
1489 rc = -nvlist_lookup_byte_array(nvbuf, XATTR_NAME_LMA, &v, &s);
1490 if (likely(rc == 0)) {
1492 lma = (struct lustre_mdt_attrs *)v;
1493 lustre_lma_swab(lma);
1494 it->mit_fid = lma->lma_self_fid;
1498 /* not a Lustre object, try next one */
1505 /* we aren't prefetching in the above loop because the number of
1506 * non-Lustre objects is very small and we will be repeating very
1507 * rare. in case we want to use this to iterate over non-Lustre
1508 * objects (i.e. when we convert regular ZFS in Lustre) it makes
1509 * sense to initiate prefetching in the loop */
1511 /* 0 - there are more items, +1 - the end */
1512 if (likely(rc == 0))
1513 osd_zfs_otable_prefetch(env, it);
1515 CDEBUG(D_OTHER, "advance: %llu -> %llu "DFID": %d\n", dnode,
1516 it->mit_pos, PFID(&it->mit_fid), rc);
1522 static struct dt_key *osd_zfs_otable_it_key(const struct lu_env *env,
1523 const struct dt_it *di)
1528 static int osd_zfs_otable_it_key_size(const struct lu_env *env,
1529 const struct dt_it *di)
1531 return sizeof(__u64);
1534 static int osd_zfs_otable_it_rec(const struct lu_env *env,
1535 const struct dt_it *di,
1536 struct dt_rec *rec, __u32 attr)
1538 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1539 struct lu_fid *fid = (struct lu_fid *)rec;
1548 static __u64 osd_zfs_otable_it_store(const struct lu_env *env,
1549 const struct dt_it *di)
1551 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1556 static int osd_zfs_otable_it_load(const struct lu_env *env,
1557 const struct dt_it *di, __u64 hash)
1559 struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
1562 it->mit_prefetched = 0;
1563 it->mit_prefetched_dnode = 0;
1565 return osd_zfs_otable_it_next(env, (struct dt_it *)di);
1568 static int osd_zfs_otable_it_key_rec(const struct lu_env *env,
1569 const struct dt_it *di, void *key_rec)
1574 const struct dt_index_operations osd_zfs_otable_ops = {
1576 .init = osd_zfs_otable_it_init,
1577 .fini = osd_zfs_otable_it_fini,
1578 .get = osd_zfs_otable_it_get,
1579 .put = osd_zfs_otable_it_put,
1580 .next = osd_zfs_otable_it_next,
1581 .key = osd_zfs_otable_it_key,
1582 .key_size = osd_zfs_otable_it_key_size,
1583 .rec = osd_zfs_otable_it_rec,
1584 .store = osd_zfs_otable_it_store,
1585 .load = osd_zfs_otable_it_load,
1586 .key_rec = osd_zfs_otable_it_key_rec,
1590 int osd_index_try(const struct lu_env *env, struct dt_object *dt,
1591 const struct dt_index_features *feat)
1593 struct osd_object *obj = osd_dt_obj(dt);
1596 LASSERT(dt_object_exists(dt));
1599 * XXX: implement support for fixed-size keys sorted with natural
1600 * numerical way (not using internal hash value)
1602 if (feat->dif_flags & DT_IND_RANGE)
1605 if (unlikely(feat == &dt_otable_features)) {
1606 dt->do_index_ops = &osd_zfs_otable_ops;
1610 LASSERT(obj->oo_db != NULL);
1611 if (likely(feat == &dt_directory_features)) {
1612 if (osd_object_is_zap(obj->oo_db))
1613 dt->do_index_ops = &osd_dir_ops;
1616 } else if (unlikely(feat == &dt_acct_features)) {
1617 LASSERT(fid_is_acct(lu_object_fid(&dt->do_lu)));
1618 dt->do_index_ops = &osd_acct_index_ops;
1619 } else if (osd_object_is_zap(obj->oo_db) &&
1620 dt->do_index_ops == NULL) {
1621 /* For index file, we don't support variable key & record sizes
1622 * and the key has to be unique */
1623 if ((feat->dif_flags & ~DT_IND_UPDATE) != 0)
1626 if (feat->dif_keysize_max > ZAP_MAXNAMELEN)
1628 if (feat->dif_keysize_max != feat->dif_keysize_min)
1631 /* As for the record size, it should be a multiple of 8 bytes
1632 * and smaller than the maximum value length supported by ZAP.
1634 if (feat->dif_recsize_max > ZAP_MAXVALUELEN)
1636 if (feat->dif_recsize_max != feat->dif_recsize_min)
1639 obj->oo_keysize = feat->dif_keysize_max;
1640 obj->oo_recsize = feat->dif_recsize_max;
1641 obj->oo_recusize = 1;
1643 /* ZFS prefers to work with array of 64bits */
1644 if ((obj->oo_recsize & 7) == 0) {
1645 obj->oo_recsize >>= 3;
1646 obj->oo_recusize = 8;
1648 dt->do_index_ops = &osd_index_ops;